Page 204 - Demo
P. 204
202Chapter12attentionmapsorclass-activationmapsderived fromdeep-learningmodelscouldalsogiveadegreeoraquantification ThereforetheidealfuturemethodconsistsofadeeplearningmethodwithcomprehensivefeaturesThisprovidesarobustmethodthatcanbeusedforbothclassificationandquantificationofcranialdeformationsFigure2AdditionalgearswillenablethecompletegearsystemthatrepresentstheclinicalworkflowtoworkmoreefficientNEWDEVELOPMENTSINACQUIRING3DPHOTOSAboveallmaking3DphotogrammetrymoreaccessibleathomeoratotherclinicalsiteswillbeamassiveimprovementAllourstudiesbenefitedfromanexcellentbutratherexpensive3DMDcranialphotosystemas wellas frompropertrainedphotographersandtechnicalmedicinestudentswhoarealwayswillingtomake3DphotosUnfortunatelytheseadvantagesarenotalwaysandeverywhereavailableNewdevelopmentssuchasthefullyautomatedsmartphone-based3Dcapturingsystemthatusesaskincapwithmarkerscouldsolvethisproblem67Othermethodsthatuse2Dphotosfromdifferentanglesandamachine-learningapproacharealsopromising8Ifa3Dphotocanbecheapandeasilyacquiredinsufficientqualityforourclassificationandanalysistoolswecoulddiagnosepatientsearlierandmakefollow-upmoreconvenient
